Net primary productivity (NPP) is a critical indicator for evaluating the carbon sequestration potential of an ecosystem and regional sustainable development, as its spatiotemporal dynamics are jointly influenced by natural and anthropogenic factors. This study investigated the Sangong River Basin, an inland watershed located in northwestern China. By employing the Carnegie–Ames–Stanford Approach (CASA) model and the Geodetector method, integrated with remote sensing data and field surveys, we systematically analyzed the spatiotemporal evolution and driving mechanisms of NPP from 1990 to 2020. Our results reveal an average annual basin-wide NPP increase of 2.33 g C·m⁻²·a⁻¹, with plains experiencing significantly greater increases (2.86 g C·m⁻²·a⁻¹) than mountains (1.71 g C·m⁻²·a⁻¹). Land use intensity (LUI) explained 31.44% of the NPP variability in the plains, whereas climatic factors, particularly temperature (71.27% contribution rate), primarily governed the NPP dynamics in mountains. Soil properties exhibited strong associations with NPP. Specifically, a 1 g·kg⁻¹ increase in soil organic content elevated NPP by 99.04 g C·m⁻²·a⁻¹, while a comparable rise in soil salinity reduced NPP by 123.59 g C·m⁻²·a⁻¹. These findings offer spatially explicit guidance for ecological restoration and carbon management in arid inland basins, underscoring the need for a strategic equilibrium between agricultural intensification and ecosystem conservation to advance carbon neutrality objectives.